Technology ❯ Artificial Intelligence

Model Evaluation

Performance Metrics Benchmarking Benchmarks Performance Comparison Performance Benchmarking Performance Benchmarks Performance Testing Benchmark Testing Performance Improvement Safety Assessments

OpenAI Publishes Framework Showing Chain‑of‑Thought Monitoring Bests Output‑Only Oversight

The authors say higher reasoning effort can buy safer oversight at extra inference cost.

OpenAI Tests 'Confession' Channel to Get AI Models to Admit Misbehavior

Microsoft Open-Sources Fara-7B, an On-Device Computer-Use Agent Built for Screenshots

Allen Institute for AI Releases Olmo 3 With Full Model-Flow Transparency

xAI Rolls Out Grok 4.1 Free to All Users, Citing Faster, More Accurate Replies

Two New RAG Preprints Propose Theme-Aligned Hypergraphs and Hierarchical Agents

OpenAI Tests ChatGPT Group Chats in Four Countries, Powered by GPT-5.1

New Research Maps LLM Strengths and Failures, Advancing Practical Fixes for Reliability, Privacy and Cost

New Papers Flag Retrieval as RAG’s Weak Link, Propose Practical Fixes

Anthropic Releases Claude Haiku 4.5, Making Near‑Frontier Performance the Free Default

Enterprises Scale RAG in Production as New Studies Expose Recall Gaps and Offer Targeted Fixes

Anthropic-Led Study Finds 250 Poisoned Files Can Backdoor Large AI Models

Samsung’s Tiny Recursive Model Tops Bigger AIs on Tough Reasoning Benchmarks

Anthropic-Led Study Finds About 250 Poisoned Documents Can Backdoor LLMs Regardless of Size

OpenAI Releases GDPval Work Benchmark as Claude Leads Early Results

DeepSeek Releases Open-Source V3.2‑Exp With Sparse Attention, Cuts API Prices as Huawei Offers 0‑Day Support

Meta Releases Open-Weights Code World Model With Built-In Execution Simulation

DeepSeek Releases V3.1‑Terminus With Reliability Fixes, Mode Limits and Agent Gains

Google Releases VaultGemma, an Open 1B-Parameter LLM Trained With Differential Privacy

OpenAI Says Incentives Drive AI Hallucinations, Calls for Scoreboard Overhaul